Anthropic AI Model Uncovers Vulnerabilities in Classified U.S. Government Systems During Security Test

Anthropic’s advanced artificial intelligence model identified vulnerabilities in classified U.S. government computer systems during a recent security testing exercise, according to a report by the Associated Press citing a U.S. official.

The AI company collaborated with U.S. intelligence agencies to evaluate the capabilities of its Mythos AI model in detecting cybersecurity weaknesses. During the assessment, the model reportedly discovered several system vulnerabilities within hours. However, the AI did not attempt to exploit the flaws during the testing period, highlighting its role as a defensive security tool rather than an offensive system.

The exercise was conducted under Project Glasswing, an Anthropic-led initiative focused on assessing and mitigating risks associated with powerful artificial intelligence technologies. The program brings together government agencies and technology partners to strengthen the security of critical software and infrastructure that could be affected by increasingly capable AI systems.

Project Glasswing is designed to identify potential threats posed by advanced AI models to public safety, national security, and economic stability. By proactively testing government systems, the initiative aims to help organizations strengthen cybersecurity defenses before vulnerabilities can be exploited by malicious actors.

The report also comes amid growing tensions between Anthropic and the Trump administration over the regulation and deployment of artificial intelligence technologies. Anthropic has previously expressed concerns regarding the military application of its AI models, while the administration has reportedly imposed restrictions on the use of certain Anthropic systems within government environments.

Earlier this month, the administration issued a directive requiring Anthropic to prevent foreign nationals from accessing its latest AI models, Fable 5 and Mythos 5. The move reflects increasing government scrutiny of advanced artificial intelligence technologies and concerns about protecting sensitive capabilities from potential foreign access.

The findings underscore the expanding role of AI in cybersecurity, as government agencies and technology companies explore how advanced models can help identify security vulnerabilities, protect critical infrastructure, and enhance national security in an increasingly digital world.